Agents in the Wild

Formula 1's agents cut data-source onboarding from 6-8 weeks to about 40 minutes of code generation

Formula 1 and AWS built agents that turn a requirements document into pull requests for data pipelines. Engineers still approve every PR, and deployment and review take extra hours.

Organization
Formula 1
Industry
Motorsport and entertainment
When
Aug 2026
Evidence
Company reported
Agents and tools
Amazon Bedrock AgentCoreClaude Sonnet 4.6GitHubJiraAWS GlueDBTAmazon SageMaker Unified Studio

Reported outcome

Data-source onboarding fell from 6-8 weeks to about 40 minutes of code generation plus hours of review; the post says agents do 95% of onboarding tasks unaided.

  • 6-8 weeks to ~40 min of code generation plus hours of review (company/AWS)
    Onboarding time
  • 95% (company/AWS)
    Tasks handled without human help
  • 18 months cleared in weeks (company/AWS)
    Backlog
  • -50% (company/AWS)
    Identity resolution processing time

Caveats

Stated by the company that uses the agent (blog, earnings call, interview). Not independently audited.

  • The post is co-written by AWS and Formula 1 staff and promotes AWS products. No independent check is given.
  • The 40 minutes covers code generation only. Deployment and review take hours more, and every pull request needs engineer approval.
  • The 95% autonomy figure is not defined: the post does not say what counts as a task or how it was measured.
  • The post gives the old baseline as '6 to 8 weeks' in places and 'up to 8 weeks' in others.
  • The authors say earlier platform work also helped. The result reflects F1's own platform and team.

What the agent does

Given a Business Requirements Document, an agent writes the configuration files for a new data source and opens a pull request. After an engineer approves it, the agent opens three more: infrastructure code (AWS Glue), DBT transformations and governance policies. All link to one Jira ticket. Other agents tag each column as personal, sensitive or pseudonymized data for GDPR, and detect upstream schema changes and propose code updates.

What happened

An S3 upload starts a Lambda function, which calls an agent on Bedrock AgentCore. The agent uses modular skills (schema mapping, data quality, governance) and reaches GitHub and Jira through scoped credentials. The post reports that one developer took the system from proof of concept to production in 4 months.

What to copy

  • Make the agent open pull requests, not change systems directly. Humans approve each step.
  • Give the agent small, named skills for each job.
  • Link every change to one ticket so reviewers see the full trail.

Starter prompt

Written by us, not by Formula 1.

Here is a requirements document for a new data source: {brd}
Here are our naming and governance rules: {rules}

Draft a configuration file for the new source.
List every column and tag it: personal, sensitive, or none. Give a reason for each tag.
List what you are unsure about. Do not guess.

Sources

  1. From weeks to minutes: How Formula 1 uses agentic AI on AWS to accelerate data operations — AWS Machine Learning Blog, Aug 3, 2026

Added Oct 26, 2026. Spotted an error? Send a correction.

Get new use cases weekly

One short email. Sources included.

Related use cases